Select a calendar:
Filter April Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Events for April 27, 2017
-
PhD Defense - Chien-Chun Hung
Thu, Apr 27, 2017 @ 02:15 AM - 04:15 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate:
Chien-Chun Hung
Title:
Resource Scheduling in Geo-distributed Computing
Date & Time:
April 27th, Thursday; 2:15-4:15pm
Room:
SAL 322
Committee:
Professor Leana Golubchik (advisor)
Professor Bhaskar Krishnamachari (external member)
Professor Wyatt Lloyd
Professor Minlan Yu
Doctor Ganesh Ananthanarayanan (Microsoft Research)
Abstract:
Due to the growing needs in computing and the increasing volume of data, cloud service providers deploy multiple datacenters around the world in order to provide fast computing response. Many applications utilizing such geo-distributed deployment include web search, user behavior analysis, machine learning applications and live camera feeds processing. Depending on the characteristics of the applications, their data may be generated, stored, and processed across the geo-distributed sites. Hence, how to efficiently process the data across the geo-distributed sites has become critical for the applications' performance.
Existing solutions first aggregate all the required data to one location and execute the computation within the site. Such solutions incur a large amount of data transfer across the WAN and lead to prolonged response time for the applications due to the significant network delay. An emerging trend is to instead distribute the computation across the sites based on data distribution, and aggregate only the results afterward. Recent works have shown such new approach results in an improvement of 3-19X in response time, or 250X in the reduction of WAN bandwidth usage.
Despite the preliminary gains, the performance of the geo-distributed jobs highly depends on how the resources are scheduled, which raises new challenges as the trivial extensions of state-of-the-art scheduling solutions lead to sub-optimal performance.
In this thesis, we first take an initiative step for improving the performance of geo-distributed jobs from the perspective of computation resource. We provide the insights into how conventional Shortest Remaining Processing Time (SRPT) falls short due to the lack of scheduling coordination among the sites, and propose a light-weight heuristic that significantly improves the jobs' response time. We also design a new job scheduling heuristic that coordinates the workload demands and the resource availability among the sites, and greedily schedule for the job that can quickly finish.
The trace-driven simulation studies show that our proposed scheduling heuristics effectively reduce the response time for the geo-distributed jobs by up to 50%.
Next, we take a step further by addressing the geo-distributed jobs' performance from the perspectives of both the computation and the network resources. Specifically, we address the scheduling challenge of the heterogeneity of the resources availability across the sites and the mismatch of the data distribution across the geo-distributed sites. We formulate the task placement decisions into Linear Programming optimization, and allocate the resources to the job that can finish quickly. In addition to the response time, our design can also nicely incorporate other performance goals, e.g., fairness and WAN usage, with simple control knobs. The EC2-based deployment of our prototype and the large-scale trace-driven simulations showed that our solutions can improve the response time of the baseline in-place scheduling approach by up to 77%, and improve the state-of-the-art geo-distributed analytics solution by up to 55%.
Finally, we expand to a more general setting in which each job has multiple configuration options, and its quality depends on the configuration it utilizes. We motivate this problem by the scenario of processing live camera feeds across hierarchical clusters. In this setting, we focus on the scheduling problem of jointly deciding job configuration and placement for concurrent jobs, and design efficient heuristic to maximize the overall quality with available resources across the geo-distributed sites. Our evaluation based on the Azure deployment of our prototype showed that the proposed solution outperforms the stat-of-the-art video analytics scheduler by up to $2.3X$, and outperforms the widely deployed Fair Scheduler by up to $15.7X$, in terms of the average quality of the concurrent jobs.
Location: Henry Salvatori Computer Science Center (SAL) - 322
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
CS Colloquium and RASC seminar: Steven Waslander (University of Waterloo) - Gimballed multi-camera localization and mapping for aerial vehicles
Thu, Apr 27, 2017 @ 11:00 AM - 12:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Steven Waslander, University of Waterloo
Talk Title: Gimballed multi-camera localization and mapping for aerial vehicles
Series: RASC Seminar Series
Abstract: This lecture satisfies requirements for CSCI 591: Computer Science Research Colloquium.
Multi-camera clusters used for visual SLAM assume a fixed calibration between the cameras, which places many limitations on its performance, and directly excludes all configurations where a camera in the cluster is mounted to a moving component. We present a calibration method and SLAM solution for dynamic multi-camera clusters, where one or more of the cluster cameras is mounted to an actuated mechanism, such as a gimbal or robotic manipulator. Our approach parametrizes the actuated mechanism using the Denavit-Hartenberg convention, then determines the calibration parameters which allow for the estimation of the time varying extrinsic transformations between camera frames. We rely on joint encoder data or camera-attached IMU to identify the extrinsic transformations during operation, and are developing active calibration methods to automate the process in the field. We validate our calibration approach using a dynamic camera cluster consisting of a static camera and a camera mounted to a pan-tilt unit as well as on a four-camera system with a single three-axis gimballed unit on a hexacopter aerial vehicle, and demonstrate that dynamic camera clusters can be provide accurate pose tracking when used to perform SLAM.
Biography: Prof. Steven Waslander is an Associate Professor in the Department of Mechanical and Mechatronics Engineering at the University of Waterloo in Waterloo, Ontario, Canada and director of the Waterloo Autonomous Vehicles Laboratory (WAVELab, http://wavelab.uwaterloo.ca). He received his B.Sc.E.in 1998 from Queen's University, his M.S. in 2002 and his Ph.D. in 2007, both from Stanford University in Aeronautics and Astronautics. He is the Program Co-Chair for the CIPPRS Computer and Robot Vision Conference, the Competition Chair for the IEEE/RSJ International Conference on Intelligent Robots and Systems and the former General Chair of the International Autonomous Robot Racing competition. His research interests lie in the areas of autonomous aerial and ground vehicles, autonomous driving, simultaneous localization and mapping, quadrotor vehicles, and machine learning. Prof. Waslander currently collaborates with numerous industrial partners, including Aeryon Labs, Clearpath Robotics, Nuvation Engineering, Denso, Renesas Electronics Corp, Qnx, and Applanix, and is a member of the NSERC Canadian Field Robotics Network. He also acts as the academic advisor to the University of Waterloo Robotics Team, which compete in multiple competitions, including the NASA Sample Return Robot Challenge, the Intelligent Ground Vehicle Competition and the University Rover Challenge.
Host: Gaurav Sukhatme
Location: Ronald Tutor Hall of Engineering (RTH) - 217
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
Measurement and Analysis of Mobile and Social Networks
Thu, Apr 27, 2017 @ 11:00 AM - 12:15 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Athina Markopoulou, Professor/UC Irvine
Talk Title: Measurement and Analysis of Mobile and Social Networks
Abstract: The majority of Internet traffic today is through mobile devices and social media. Large-scale measurement and analysis of these systems is necessary in order to understand underlying patterns and enable engineering optimizations and new applications. In this talk, I will present highlights of our research in this area.
First, I will discuss online social networks. I will present our "2K+" framework for generating synthetic graphs that resemble online social networks, in terms of joint degree distribution and additional characteristics, such as clustering and node attributes [INFOCOM'13, INFOCOM'15]. This problem was motivated by our prior work on graph sampling [JSAC'11, SIGMETRICS'11, INFOCOM'10] and by popular demand to make the Facebook datasets we collected publicly available.
Second, I will discuss cellular networks. I will present our work on analyzing Call Detail Records (CDRs) in order to characterize human activity in urban environments, with applications to urban ecology [MOBIHOC'15] and ride-sharing [UBICOMP'14, SIGSPATIAL'15-16].
Third, I will present our ongoing work on AntMonitor - a system for monitoring network traffic on mobile devices [SIGCOMM C2BID'15], with applications to privacy leaks detection [MOBICOM Demo'15], crowdsourcing of network performance measurements, and improved wireless access.
Biography: Athina Markopoulou is an Associate Professor in EECS at the University of California, Irvine. She received the Diploma degree in Electrical and Computer Engineering from the National Technical University of Athens, Greece, in 1996, and the Master's and Ph.D. degrees in Electrical Engineering from Stanford University, in 1998 and 2003, respectively. She has held short-term/visiting appointments at SprintLabs (2003), Arista Networks (2005), IT University of Copenhagen (2012-2013), and she co-founded Shoelace Wireless (2012). She has received the NSF CAREER Award (2008), the Henry Samueli School of Engineering Faculty Midcareer Award for Research (2014), and the OCEC Educator Award (2017). She has been an Associate Editor for IEEE/ACM Transactions on Networking (2013-2015), an Associate Editor for ACM CCR (2016), the General Co-Chair for ACM CoNEXT 2016, and the Director of the Networked Systems program at UCI. Her research interests are in the area of networking including mobile systems and mobile data analytics, network measurement, online social networks, network security and privacy, network coding, and multimedia traffic.
Host: Professor Konstantinos Psounis, kpsounis@usc.edu
More Information: Seminar Announcement - Markopoulou 042717.pdf
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 132
Audiences: Everyone Is Invited
Contact: Mayumi Thrasher
-
Le Val Lund Lecture with Student Symposium
Thu, Apr 27, 2017 @ 01:00 PM - 09:00 PM
Sonny Astani Department of Civil and Environmental Engineering
Conferences, Lectures, & Seminars
Speaker: Craig Davis, Technical Speaker and Recipient of the 2016 ASCE Le Val Lund Award for Practicing Lifeline Risk Reduction
Talk Title: Operationalizing Resilience for Lifeline Systems
Host: ASCE
More Information: Final_LeVal Lund_Lecture_save_the_date_v4_27April17.pdf
Location: California Institute of Technology
Audiences: Everyone Is Invited
Contact: Kaela Berry
-
PhD Defense
Thu, Apr 27, 2017 @ 01:00 PM - 03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Akshay Gadde, University of Southern California
Talk Title: Sampling and Filtering of Signals on Graphs with Applications to Active Learning and Image Processing
Abstract: Processing of signals defined over the nodes of a graph has generated a lot of interest recently. This is due to the emergence of modern application domains such as social networks, web information analysis, sensor networks and machine learning, in which graphs provide a natural representation for the data. Traditional data such as images and videos can also be represented as signals on graphs. A frequency domain representation for graph signals can be obtained using the eigenvectors and eigenvalues of operators which measure the variation in signals taking into account the underlying connectivity in the graph. Spectral filtering can then be defined in this frequency domain. Based on this, we develop a sampling theory for graph signals by answering the following questions: 1. When can we uniquely and stably reconstruct a bandlimited graph signal from its samples on a subset of the nodes? 2. What is the best subset of nodes for sampling a signal so that the resulting bandlimited reconstruction is most stable? 3. How to compute a bandlimited reconstruction efficiently from a subset of samples? The algorithms developed for sampling set selection and reconstruction do not require explicit eigenvalue decomposition of the variation operator and admit efficient, localized implementation. Using graph sampling theory, we propose effective graph based active semi-supervised learning techniques. We also give a probabilistic interpretation for the proposed techniques. Based on this interpretation, we generalize the framework of active learning on graphs using Bayesian methods to give an adaptive sampling method. Additionally, we study the application graph spectral filtering in image processing by representing the image as a graph, where the nodes correspond to the pixels and edge weights capture the similarity between them given by the coefficients of the bilateral filter. We show that the bilateral filter is a low pass graph spectral filter with linearly decaying spectral response. We then generalize the bilateral filter by defining filters on the above graph with different spectral responses depending on the application. We also consider the problem of constructing a sparse graph from the given data efficiently, which can be used in graph based learning and fast image adaptive filtering.
Biography: Akshay Gadde received his Bachelor of Technology degree in Electrical Engineering from Indian Institute of Technology (IIT), Kharagpur, India in 2011. He has been working towards a Ph.D. in Electrical Engineering at the University of Southern California (USC), Los Angeles since 2011. His work (with Prof. Antonio Ortega and Aamir Anis) won the Best Student Paper Award at ICASSP 2014. His research interests include graph signal processing and machine learning with applications to multimedia data processing and compression.
Host: Dr. Antonio Ortega
More Information: Gadde Seminar Announcement.png
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Gloria Halfacre
-
Computer Architectures for Deep Learning Applications
Thu, Apr 27, 2017 @ 03:30 PM - 05:30 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: David Brooks, Harvard University
Talk Title: Computer Architectures for Deep Learning Applications
Abstract: Deep learning has been popularized by its recent successes on challenging artificial intelligence problems. One of the reasons for its dominance is also an ongoing challenge: the need for immense amounts of computational power. Hardware architects have responded by proposing a wide array of promising ideas, but to date, the majority of the work has focused on specific algorithms in somewhat narrow application domains. While their specificity does not diminish these approaches, there is a clear need for more flexible solutions. We believe the first step is to examine the characteristics of cutting edge models from across the deep learning community. Consequently, we have assembled Fathom: a collection of eight archetypal deep learning workloads for study. Each of these models comes from a seminal work in the deep learning community, ranging from the familiar deep convolutional neural network of Krizhevsky et al., to the more exotic memory networks from Facebook's AI research group. Fathom has been released online, and this talk describes the fundamental performance characteristics of each model. We use a set of application-level modeling tools built around the TensorFlow deep learning framework in order to analyze the behavior of the Fathom workloads. We present a breakdown of where time is spent, the similarities between the performance profiles of our models, an analysis of behavior in inference and training, and the effects of parallelism on scaling. The talk will then consider novel computer architectures that can improve the performance and efficiency of deep learning workloads.
Biography: David Brooks is the Haley Family Professor of Computer Science in the School of Engineering and Applied Sciences at Harvard University. Prior to joining Harvard, he was a research staff member at IBM T.J. Watson Research Center. Prof. Brooks received his BS in Electrical Engineering at the University of Southern California and MA and PhD degrees in Electrical Engineering at Princeton University. His research interests include resilient and power-efficient computer hardware and software design for high-performance and embedded systems. Prof. Brooks is a Fellow of the IEEE and has received several honors and awards including the ACM Maurice Wilkes Award, ISCA Influential Paper Award, NSF CAREER award, IBM Faculty Partnership Award, and DARPA Young Faculty Award.
Host: Xuehai Qian, x04459, xuehai.qian@usc.edu
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Gerrielyn Ramos
-
CS & ML Colloquium: Matus Telgarsky (UIUC) - Representation power of neural networks
Thu, Apr 27, 2017 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Matus Telgarsky, UIUC
Talk Title: Representation power of neural networks
Series: Yahoo! Labs Machine Learning Seminar Series
Abstract: This lecture satisfies requirements for CSCI 591: Computer Science Research Colloquium.
This talk will present a series of mathematical vignettes on the representation power of neural networks. Amongst old results, the classical universal approximation theorem will be presented, along with Kolmogorov's superposition theorem. Recent results will include depth hierarchies (for any choice of depth, there exists functions which can only be approximated by slightly less deep networks when they have exponential size), connections to polynomials (namely, rational functions and neural networks well-approximate each other), and the power of recurrent networks. Open problems will be sprinkled throughout.
Biography: Matus Telgarsky is an assistant professor at UIUC. He received his PhD in 2013 at UCSD under Sanjoy Dasgupta. He works in machine learning theory; his current interests are non-convex optimization and neural network representation.
Host: CS Department
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
